To circumvent the problem in GPS／INS integrated navigation for data infusion that Kalman filter degrades severely since the statistics of the noise might be time-variant, an adaptive Kalman filtering algorithm based on variational Bayesian learning is suggested and used in the integrated navigation system model in which both the moment of noise and the states are considered as stochastic parameters and estimated together. Using a probabilistic approach, a concrete derivation is given to represent how variational Bayesian learning works in a recursive way to approximate the true posterior of the noise together with the states. Experimental results demonstrate that the proposed filter is adaptive and performs well in tracking variances of the noise and estimating the states including position and velocity in GPS／INS integrated navigation system.